The Pulse
Microsoft releases seven AI models at Ignite
Microsoft's new lineup includes models for coding, cybersecurity, image editing, audio and decisions. The company says its models offer lower costs than competing systems on several tasks.

Microsoft is releasing seven AI models at Ignite, including systems for coding, cybersecurity, image editing and audio, with the company claiming lower costs than competing models on several tasks. The lineup also includes a decision model and transcription technology used in Dragon Copilot, according to a Microsoft AI post published Nov. 17.
Coding models and benchmarks
MAI Code 2 is designed for enterprise coding tasks including debugging, multi-file implementation and reasoning across a repository. MAI Code 2 Flash is designed for high-volume work such as explaining code, making targeted edits and routine fixes. Neither coding model has yet been tuned for harder problems such as fundamental science research and very long-horizon agentic work, Microsoft said.
Across four coding benchmarks, the company says MAI Code 2 achieved 98% of GPT-6.1 Sol's performance at about 40% of its price. The same comparison put it at 96% of Claude Opus 5.5's performance at 5% of the price.
Microsoft says Code 2 Flash delivered comparable performance to GPT-6 Luna at 65% of the cost and Claude Haiku 5.5 at 20% of the cost across three benchmarks for routine coding. Both coding models are available in GitHub, while Code 2 Flash is also in a limited-access preview in Microsoft Foundry.
Security and image editing
For cybersecurity, Microsoft says MAI Cyber 2 Flash matches GPT-6 Astra's quality on the CyberGymE2E benchmark at less than 2% of the cost per task. The model is built to find and validate vulnerabilities in code and help defenders investigate and detect attacks.
Verified defenders can access it through Codename MDASH, Microsoft's multi-agent security system. The system scans code for vulnerabilities, checks potential findings and can suggest and validate patches.
Microsoft says MAI Image 2.7 ranks first on Artificial Analysis for image editing and outperforms GPT Image 2.6 Sunburst at maximum settings at one fifth of the price.
Audio and decisions
MAI Realtime 1 starts at $0.01 per audio minute. Microsoft says it outperforms GPT-Realtime 2.1.
Dragon Copilot Transcribe is built on MAI Transcribe 2 and already powers 10 million patient encounters a month, with multilingual error rates halved, the company says. Microsoft also describes MAI Transcribe 2 as the world's fastest and most accurate multilingual transcription model.
Microsoft says Decision 1.1, its latest decision model, is 50 times faster and 79 times cheaper than GPT-Sol 6.1.
Sources
Related stories
- Microsoft makes Decision-1.1 generally available with image support
- Microsoft plans licensed MAI model weights through Frontier Company deals
- Microsoft to bring MAI Cyber 2 Flash to its security products
- Microsoft and Salesforce bring sales workflows to Copilot in preview
- Microsoft makes Copilot Home generally available as Autopilot rolls out